Menu

Browse all skills

Content agent skill

FAQ Generator

build an FAQ set that does two jobs. It lets real customers serve themselves and reduces support load. It also gives search and AI engines clear, self-contained, accurate answers to definitional and validation questions. Writing rules are in [references/writing-and-titling.md](references/writing-and-titling.md). Sourcing rules are in [references/evidence-standards.md](references/evidence-standards.md).

View raw SKILL.mdView package on GitHub

Important platform fact (see references/evidence-standards.md): Google no longer shows FAQ rich results. Do not promise rich results from FAQPage markup. FAQs earn their value through on-page usefulness, support deflection, and being a clear answer source.

When to use this skill

Use for FAQ pages, FAQ blocks on product or pricing pages, help-center question sets, or researching which questions customers ask.

Do not use when:

  • The user wants a full article where a FAQ is only one section. Use content, which writes the FAQ itself.
  • The user wants legal or policy text such as terms or refunds written from scratch. That needs a qualified owner. This skill can only phrase questions about existing approved policy.

Required inputs

brand:
  name: ""
  website: ""
scope: ""                   # product, page, or topic area the FAQ covers
market:
  country: ""
  language: ""
question_sources:           # provide whatever exists
  support_tickets: ""       # export, summary, or "none"
  sales_objections: ""
  search_data: ""           # PAA/autocomplete/GSC queries
  prior_findings: ""        # competitor-audit / keyword-research prompt results
fact_sources: []            # pricing pages, policy docs, product docs (authoritative answers)
voice_sample: ""

If you have no question sources and no fact sources, stop and ask. Do not invent questions or answers.

Workflow

- [ ] 1. Questions collected from real sources, each tagged with origin
- [ ] 2. Questions deduplicated, prioritized, grouped into themes
- [ ] 3. Answers written answer-first from authoritative fact sources
- [ ] 4. Placement recommended per question
- [ ] 5. Volatile answers given an owner and review date
- [ ] 6. Output assembled; unanswerable questions listed

  1. Collect. Gather questions from:

    • support tickets, chat logs and sales objections (the questions that cost the business time)
    • People Also Ask, autocomplete and Search Console queries
    • validation prompts already found by competitor-audit or keyword-research ("is X legit", "how does X compare")
    • objections customer-facing teams recognize by name

    Tag each question with its origin. Mark a question as assumption only if the user explicitly asks for proactive questions.

  2. Prioritize.

    • Score each question by frequency (ticket count or search demand, where known) multiplied by the friction a good answer removes.
    • A frequent question with a simple answer beats a rare, complicated one.
    • Deduplicate near-identical phrasings.
    • Group the questions into themes.

  3. Answer.

    • Put the direct answer in the first sentence, then 1–3 supporting sentences. Use a short numbered list only for genuine processes.
    • Every factual answer (price, limit, policy, timeline) must come from a listed fact source and carry its URL and access date.
    • If no source exists, mark the answer [NEEDS SOURCE] and leave it out of the publish set.
    • Prefer specific, checkable statements to vague reassurance.
    • Match the voice sample.

  4. Place.

    • Put broad, cross-cutting questions on the FAQ page.
    • Embed page-specific questions on the relevant product, pricing or content page.
    • Do not default every question to one page.
    • Keep each Q&A self-contained so it still makes sense when quoted alone.

  5. Freshness. For any answer that depends on something that changes (price, policy, feature, legal), record an owner and a review date. The default review cycle is 90 days unless the user specifies another.

  6. Assemble using the template.

Output template

# FAQ set: <scope> — <date>

## Theme: <name>
| # | Question | Answer | Origin | Priority | Placement | Source (URL, accessed) | Volatile? | Owner | Review by |

## Not publishable yet
| Question | Why (no source / needs policy owner / conflicting sources) |

## Notes
- Markup: optional FAQPage JSON-LD is harmless but yields no Google rich result; do not report it as a ranking or visibility tactic.

Approval gates

  • Answers about pricing, refunds, legal, security, compliance or health need sign-off from the policy owner before publishing.
  • Do not publish to a CMS or help center without explicit permission.

Stop conditions

  • There are no real question sources and the user will not supply any.
  • The authoritative sources contradict each other, for example the pricing page and the docs disagree. Report the conflict instead of choosing one.

Anti-patterns

  • Inventing a plausible-sounding question list.
  • Burying the answer under setup.
  • Flat, ungrouped lists.
  • Time-sensitive answers with no owner or review date.
  • Promising rich results or rankings from FAQ schema.
  • A generic support tone disconnected from the brand voice.

Completion checklist

  • Every question has a real, tagged origin
  • Prioritized, deduplicated and grouped into themes
  • Every answer leads with the answer, and every factual answer has a source and access date
  • Placement recommended per question
  • Volatile answers have an owner and a review date
  • Unsourced or conflicting items listed separately, not published